Enhancing the Forecast of Ocean Physical Variables through Physics Informed Machine Learning in the Santos Estuary, Brazil

Autor(es)

Moreno, Felipe M. ; Schiaveto Neto, Luiz A. ; Cozman, Fabio G. ; Dottori, Marcelo ; Tannuri, Eduardo A. ; IEEE

• Tipo de documento: Artigo Científico
• Fonte: OCEANS 2022; v. N/A, p. 7-pg., 2022-01-01.

Resumo

This work aims to improve the forecast of surface currents in the entrance of the Santos estuary in Brazil by applying Quantile Regression Forests (QRF) to estimate the error of the Santos Operational Forecasting System (SOFS), a physics-based numerical model for the region. This was achieved by using in-situ data, measured between 2019 and 2021, associated with historical forecasted data from the SOFS. The use of QRF to correct the SOFS forecasts led to a increase in skill of 0.332 in Mean Absolute Error (MAE) and almost eliminated the bias error of the predicted currents. (AU)

• Processo FAPESP: 19/07665-4 - Centro de Inteligência Artificial
• Beneficiário: Fabio Gagliardi Cozman
• Modalidade de apoio: Auxílio à Pesquisa - Programa eScience e Data Science - Centros de Pesquisa em Engenharia
• Processo FAPESP: 20/16746-5 - Physics-informed machine learning aplicado para previsões de condições metoceânicas
• Beneficiário: Felipe Marino Moreno
• Modalidade de apoio: Bolsas no Brasil - Doutorado